Study On The Error Analysis In Chinese-to-English NMT Translation

Neural network machine translation(NMT)is widely regarded as a disruptive breakthrough in the field of machine translation.It has been widely used to relieve professional translators from the burden of repetitive work by offering a draft translation.In 2014,a paper was published claiming that there has been a “technological trend” in translation studies.However,a commonly expressed concern is whether a machine can be as effective as a human translator.Numerous studies have been conducted to answer this question with regard to the evaluation of output.The measures,however,were mainly at the technical and automatic level than semantic ones.Moreover,some studies also tend to be partial and one-sided due to insufficient understanding of machine translation.This thesis starts with reviewing the history and current status of machine translation in a bid to have a comprehensive understanding of machine translation.The development of studies on the evaluation of machine translation quality and the main challenges facing NMT will also be explored.Then,from the perspective of semantics,the error analysis approach is adopted to identify,classify and analyze the errors that occurred in translations provided by three NMT systems,which represent the highest level of their kind.The source texts are three Chinese texts under topics including culture,sports,and science and technology.Based on the frequency of occurrence,errors are divided into three categories: vocabulary,sentence,and text,and then subdivided into nineteen subcategories.Through the qualitative and quantitative analysis,findings show that NMT systems have the worst performance in rendering words,in particular,terminologies and idioms.At the sentence level,the main mistakes include incorrect segmentation,missing sentence components,and so on.While at the text level,the lack of logic and cohesion turns out to be one of the major difficulties.The experiment results also reveal that frequent errors can be attributed to the significant difference between Chinese and English.As a result,further studies are required regarding the differences between the two languages and how better translation can be achieved.